Regression Results

Proxy Voting Recommendations and Voting Outcomes

Table 2 - 9 provides regression results. In line with the paper’s first prediction and consistent with the descriptive results discussed above, Model 1 shows that negative ISS recommendations correlate with 8.5% less supportive shareholder votes. This drop is even

72.5 69.8 55.7

82.2

50 61.5

44.3

76.6 77.1

58.9 85.1

53.9 61.8

50.6

5075100

Against For

(1) (2) (3) (4) (5) (6) (7) (1) (2) (3) (4) (5) (6) (7)

(1) Mean of GRId (in %) (2) Mean of GRId Board Subscore (in %)

(3) Mean of GRId Compensation Subscore (in %) (4) Mean of GRId Shareholder Rights / Audit Subscore (in %) (5) Mean of Firm's Total Assets (in %) (6) Mean of Firm's Free Float (in %)

(7) Mean of Firm's HDAX Membership (in %)

Firm Characteristics (mean values in %)

Graphs by ISS Rec. ('Against Rec.' vs. 'For Rec.')

LF-Report Sample: 918 voting items based on 92 firms

ISS Recommendations and Firm Characteristics

38 more pronounced when considering voting items with high client base (11.21%), low voting turnout (11.59%), and high free float (11.44%).³⁹

Table 2 - 9

In addition, when examining the subsample of firms which are covered by ISS LF-reports (i.e., firms with above-average client base), ISS recommendations correlate with 16.32% and 16.11% less supportive shareholder votes for voting items with below-average turnout and above average free float, respectively (Models 5 and 6, Panel A, ).⁴⁰

Table 2 - 9: Regression Results: Prediction 1 – Firm-Fixed-Effects Regressions

These results indicate that ISS voting recommendations significantly correlate with shareholder votes on a statistically as well as economically meaningful level. Overall, they suggest that – despite differences in the institutional arrangements between the U.S. and Germany – proxy voting advisors might play an influential role at German AGMs as well. In economic terms, however, ISS voting recommendations correlate with voting outcomes at a comparably lower level (8.5% to 19%

and 26% as documented by Cai et al., 2009 and Ertimur et al., 2013, respectively).

Pred.

Sign

Dependent Variable: VOTING RESULT (in %) Full Sample

(LF- and SF- ISS reports) LF-Sample

(N with high Client Base)

Linear Prediction of Voting Result (in %) if

(a) ISS=1 & Moderator=0 93.75 92.44 92.25 90.29 90.73

(b) ISS=1 & Moderator=1 87.16 86.68 86.92 81.92 82.19

Notes: Underlying regression model is:

VOTING_RESULTiv = α+γ₁ISS_AIGAINSTiv+γ₂MODERATORiv+γ₃ISS_AGIANST × MODERATORiv+ε The dependent variable VOTING_RESULTiv stands for the voting result (in %) casted in favor of a specific voting item of firm i and AGM voting item v. ISS_AGAINST is a variable indicating with 1 if ISS recommends to vote against a specific AGM item, and zero otherwise. MODERATOR stands for different variables which are expected to moderate the relationship between ISS “vote against” recommendations and voting results, i.e., FREE FLOAT (with one if firm’s free float is above average, and zero otherwise), invTURNOUT (with one if firm’s voting presence is below average, and zero otherwise), and CLIENT BASE (with one if firm is covered by ISS LF-report, and zero otherwise). The regression models have robust standard errors which are one-way clustered at AGM voting item level. To control for (observed / unobserved) firm characteristics the regression models contain firm-fixed effects. For detailed descriptions of the variables, see Appendix 2 - 1 . Reported values: coefficient (t-value) *** (**) (*) indicates a significance level at 1% (5%) (10%), two-tailed.

39 Except for invTURNOUT, firm fixed-effects capture the main effects of FREE FLOAT and CLIENT BASE.

40 To shed further light on the potential moderating effects of client base, free float, and voting turnout, I extend the basic regression model with twofold interaction terms. Untabulated results reveal that ISS voting recommendations correlate with 18.56% less supportive shareholder votes for voting items with high client base, low voting turnout, and high free float.

39 Proxy Voting Recommendations and Governance Ratings

Table 2 - 10 addresses the paper’s second prediction and provides the results of the corresponding probit regressions. Across all different models, ISS governance ratings (GRId) significantly correlate with ISS decisions to issue negative voting recommendations. In contrast to this, two out of three control variables, i.e., ownership structure and blue chip index membership, remain insignificant across most models. Besides the statistical significance (p-values are consistently below 1%), the GRId correlations are economically meaningful. For example, an increase from the lowest to the highest rated firm (an increase from 5 to 12 in the GRId) reduces the probability of receiving a negative ISS recommendation by more than 20 percentage points (Model 1, Table 2 - 10).⁴¹

Table 2 - 10

This is even more pronounced – with a reduction of over 50 percentage points – when considering a subsample of only non-routine voting items (Model 2, ). In addition, Model 3 and 4 (Model 5 and 6) provide corresponding regression results on a subsample of voting items with respect to board elections (compensation issues). If ISS consistently evaluates the board quality as well as the quality of the remuneration system across its voting recommendations and its commercially available governance ratings, one might expect significant correlations especially for the respective subratings GRId_BOARD and GRId_COMP.

Consistent with this, Model 4 (Model 6) shows that ISS recommendations against the election of supervisory board members (against compensation issues) are significantly correlated with ISS’s evaluations of the corresponding board quality (the remuneration system’s quality). For example, the predicted probability of receiving a “vote against”

recommendation by ISS on director election (compensation) proposals is 58.19% (66.98%) and 8.63% (1.29%) for firms with the lowest and the highest board (compensation) score, respectively.⁴²

One reason for the divergent results might rest upon the different time frames and the different employed ISS governance ratings. In contrast to the U.S. findings provided by Daines et al. (2010) which are based on ISS CGQ ratings, this study employs ISS GRId governance ratings. From 2002 until 2009, ISS’s governance ratings were marketed as Overall, these findings contrast the U.S. results provided by Daines et al.

(2010) and suggest that the employed governance perception of ISS is potentially consistent across its different commercially available products.

41 The final rating score, GRId score, ranges theoretically (empirically for my sample) between 0 (5) and 12 (12).

Higher GRId scores indicate better governance quality.

42 Both scores have numbers between 1 and 3, with higher scores reflecting better governance.

40 Corporate Governance Quotient (CGQ). In 2010, ISS re-launched the rating under the name Governance Risk Indicator (GRId). However, Larcker and Tayan (2011, p. 440) note that the GRId rating is not materially different to the CGQ rating. Nevertheless, the alignment between both, the methodology of ISS governance ratings and the underlying principles of ISS proxy voting policies, might have increased after 2010. Although ISS already attributes a general alignment between its governance ratings and its proxy voting guidelines prior to the re-launch in 2010 (ISS, 2007, p. 24), it explicitly highlights the alignment between both products afterwards. In particular, ISS / RiskMetrics Group (2010, p. 7) states that “GRId’s methodology for assessing risk is closely aligned with the principles underlying RiskMetrics’

benchmark proxy voting guidelines”.⁴³

Table 2 - 10: Regression Results: Prediction 2 – Probit Regressions

In addition, it (2010, p. 7) further outlines that this alignment “will help [to] shape GRId, ensuing it is up-to-date, relevant, and tailored to address variations in governance practices across global capital markets”.

Pred.

43 Until the acquisition by MSCI in 2010, ISS was a subsidiary of RiskMetrics Group. In April 2014, ISS was acquired by Vestar Capital Partners.

41

The dependent variable ISS AGAINST is a dummy variable indicating with 1 if ISS recommends to vote against a specific voting item, and 0 otherwise. GRId (Governance Risk Indicator) stands for ISS’s commercial corporate governance rating. GRId_SUBSCORES reflects the different subscores of the commercial corporate governance rating (i.e., subscores for board, compensation, shareholder rights, and audits). FIRM_CONTROL is a vector of different firm-level control variables (log of total assets, free float, and blue chip (HDAX) index membership). IND stands for industry-fixed effects. The regression models have standard errors which are heteroskedasticity robust and one-way clustered at firm level. For detailed descriptions of the variables, see Appendix 2 - 1. Reported values: coefficient (t-value) *** (**) (*) indicates a significance level at 1% (5%) (10%), two-tailed.

2.5.4 Additional Analyses